Merge branch 'master' of https://gitlab.inria.fr/qdufour/donar

Dockerfile

@@ -44,14 +44,17 @@ RUN mkdir out && \
 FROM fedora:30
 RUN dnf install -y \
   glib2 \
+  procps-ng \
   valgrind \
   nmap-ncat \
   psmisc \
   libevent \
   zlib \
+  strace \
   openssl \
   libzstd \
   xz-libs \
+  moreutils \
   gstreamer1 \
   gstreamer1-plugins-base \
   gstreamer1-plugins-good \
@@ -60,7 +63,7 @@ RUN dnf install -y \
 
 WORKDIR /home/donar
 RUN mkdir /home/donar/shared && mkdir /home/donar/res && chown -R 1000 /home/donar
-USER 1000
+USER 0
 ENV HOME /home/donar
 COPY --from=builder /home/donar-build/out/donar /usr/local/bin
 COPY --from=builder /home/donar-build/out/measlat /usr/local/bin

scripts/container/dup2-server

@@ -5,7 +5,7 @@ cd $5/shared
 echo -e 'authenticate ""\nsetevents extended circ circ_minor stream orconn addrmap status_general status_client guard info notice warn err newdesc descchanged ns status_server stream_bw clients_seen hs_desc network_liveness' | \
   nc --no-shutdown 127.0.0.1 9051 | \
   ts \
-> ../log/server-tor-events-stdout.log 2> ../log/server-tor-events-stderr.log
+> ../log/server-tor-events-stdout.log 2> ../log/server-tor-events-stderr.log &
 
 donar \
   -a dup2 \

scripts/container/lightning-server

@@ -4,8 +4,17 @@ cd $6/shared
 
 echo -e 'authenticate ""\nsetevents extended circ circ_minor stream orconn addrmap status_general status_client guard info notice warn err newdesc descchanged ns status_server stream_bw clients_seen hs_desc network_liveness' | \
   nc --no-shutdown 127.0.0.1 9051 | \
-  ts \
+  ts '<%b %d %H:%M:%.S>' \
-> ../log/server-tor-events-stdout.log 2> ../log/server-tor-events-stderr.log
+> ../log/server-tor-events-stdout.log 2> ../log/server-tor-events-stderr.log &
+
+strace \
+  -s 800 \
+  -f \
+  -e trace=network,epoll_wait,epoll_create,epoll_create1,epoll_ctl,epoll_pwait,close \
+  -p 1 \
+  -o /dev/stdout | \
+  ts '<%b %d %H:%M:%.S>' \
+> ../log/server-strace-tor-stdout.log 2> ../log/server-strace-tor-stderr.log &
 
 donar \
   -a lightning \
@@ -16,4 +25,14 @@ donar \
   -e 9000 \
 > ../log/server-donar-stdout.log 2> ../log/server-donar-stderr.log &
 
+sleep 3
+strace \
+  -s 800 \
+  -f \
+  -e trace=network,epoll_wait,epoll_create,epoll_create1,epoll_ctl,epoll_pwait,close \
+  -p `pidof donar` \
+  -o /dev/stdout | \
+  ts '<%b %d %H:%M:%.S>' \
+> ../log/server-strace-donar-stdout.log 2> ../log/server-strace-donar-stderr.log &
+
 measlat -h 127.0.0.1 -p 9000 -l -c $1 -i $2 -s $3 > ../log/server-measlat-stdout.log 2> ../log/server-measlat-stderr.log

src/donar_client.c

@@ -101,10 +101,10 @@ int donar_client_stream_repair(struct evt_core_ctx* ctx, struct evt_core_fdinfo*
   if (removed == 2) {
     fprintf(stdout, "[%s][donar-client] Retriggering socks5 for port %d\n", current_human_datetime (), port);
     init_socks5_client (app_ctx, pos);
-    return 0;
+    return 1;
   } else if (removed == 0) {
     fprintf(stdout, "[%s][donar-client] Socks5 has already been retriggered for port %d\n", current_human_datetime (), port);
-    return 0;
+    return 1;
   } else {
     fprintf(stderr, "[%s][donar-client] We only removed 1 link and not 2 for port %d, strange behaviour, exiting...\n", current_human_datetime (), port);
     exit(EXIT_FAILURE);

src/donar_init.c

@@ -43,6 +43,7 @@ void signal_init(struct evt_core_ctx* evts) {
   sigaddset(&mask, SIGINT);
   sigaddset(&mask, SIGQUIT);
   sigaddset(&mask, SIGTERM);
+  sigaddset(&mask, SIGPIPE);
   if (sigprocmask(SIG_BLOCK, &mask, NULL) == -1) {
     perror("sigprocmask");
     exit(EXIT_FAILURE);

src/donar_server.c

@@ -52,9 +52,29 @@ socket_create_err:
 }
 
 struct tor_ctl* ugly_global_tctl;
-int donar_server_stream_repair(struct evt_core_ctx* ctx, struct evt_core_fdinfo* fd) {
+int donar_server_stream_repair(struct evt_core_ctx* ctx, struct evt_core_fdinfo* fdinfo) {
-  fprintf(stderr, "[%s][donar-server] I am a server, I do nothing with broken streams...\n", current_human_datetime ());
+  fprintf(stdout, "[%s][donar-server] %s broke\n", current_human_datetime (), fdinfo->url);
-  tor_ctl_list_onions(ugly_global_tctl);
+
+  struct evt_core_fdinfo* fdtarget = NULL;
+  int port = url_get_port_int (fdinfo->url);
+  int pos = port - 7500, removed = 0;
+  char buffer[256];
+
+  sprintf(buffer, "tcp:read:127.0.0.1:%d", port);
+  fdtarget = evt_core_get_from_url (ctx, buffer);
+  if (fdtarget != NULL) {
+    evt_core_rm_fd(ctx, fdtarget->fd);
+    removed++;
+  }
+
+  sprintf(buffer, "tcp:write:127.0.0.1:%d", port);
+  fdtarget = evt_core_get_from_url (ctx, buffer);
+  if (fdtarget != NULL) {
+    evt_core_rm_fd(ctx, fdtarget->fd);
+    removed++;
+  }
+  printf("[%s][donar-server] removed %d links\n", current_human_datetime (), removed);
+
   return 1;
 }
 

src/evt_core.c

@@ -2,8 +2,11 @@
 
 void free_fdinfo(void* v) {
   struct evt_core_fdinfo* fdinfo = (struct evt_core_fdinfo*)v;
-  close(fdinfo->fd); // We close the file descriptor here
+  fprintf(stdout, "[%s][evt_core] Freeing fdinfo url=%s fd=%d\n", current_human_datetime (), fdinfo->url, fdinfo->fd);
-  //fprintf(stderr, "Freeing fdinfo for %s\n", fdinfo->url);
+  if (close(fdinfo->fd) != 0) { // We close the file descriptor here
+    fprintf(stderr, "[%s][evt_core] Failed to close fd for url=%s fd=%d. ",  current_human_datetime (), fdinfo->url, fdinfo->fd);
+    perror("Error");
+  }
   if (fdinfo->free_other != NULL) {
     //fprintf(stderr, "Freeing fdinfo->other for %s\n", fdinfo->url);
     fdinfo->free_other(fdinfo->other);
@@ -266,7 +269,7 @@ void evt_core_loop(struct evt_core_ctx* ctx) {
     timing_fx_start(&tfx_loop);
     if (num_fd == -1) {
       perror("Failed to epoll_wait");
-      exit(EXIT_FAILURE);
+      continue;
     }
 
     for (n = 0 ; n < num_fd; n++) {
@@ -274,8 +277,8 @@ void evt_core_loop(struct evt_core_ctx* ctx) {
       if (events[n].events & (EPOLLERR | EPOLLHUP | EPOLLRDHUP)) {
         int err_fd = events[n].data.fd;
         int evt = events[n].events;
-        if (evt & EPOLLRDHUP) fprintf(stderr, "Epoll Read Hup Event.\n");
+        if (evt & EPOLLRDHUP) fprintf(stderr, "Epoll Read Hup Event fd=%d.\n", err_fd);
-        if (evt & EPOLLHUP) fprintf(stderr, "Epoll Hup Event.\n");
+        if (evt & EPOLLHUP) fprintf(stderr, "Epoll Hup Event fd=%d.\n", err_fd);
         if (evt & EPOLLERR) {
           int error = 0;
           socklen_t errlen = sizeof(error);

src/proxy.c

@@ -31,6 +31,8 @@ int main_on_tcp_co(struct evt_core_ctx* ctx, struct evt_core_fdinfo* fdinfo) {
   sprintf(to_fdinfo.url, "tcp:write:127.0.0.1:%s", port);
   evt_core_add_fd (ctx, &to_fdinfo);
 
+  printf("[%s][proxy] Accepted a new connection on port=%s: read_fd=%d, write_fd=%d\n", current_human_datetime (), port, conn_sock1, conn_sock2);
+
   return 0;
 
 co_error:
@@ -212,7 +214,7 @@ void algo_main_init(struct evt_core_ctx* evt, struct algo_params* ap) {
 
   struct evt_core_cat tcp_read = {
     .name = "tcp-read",
-    .flags = EPOLLIN | EPOLLET | EPOLLRDHUP,
+    .flags = EPOLLIN | EPOLLET | EPOLLHUP,
     .app_ctx = ctx,
     .free_app_ctx = algo_main_destroy,
     .cb = main_on_tcp_read,
@@ -223,7 +225,7 @@ void algo_main_init(struct evt_core_ctx* evt, struct algo_params* ap) {
 
   struct evt_core_cat udp_read = {
     .name = "udp-read",
-    .flags = EPOLLIN | EPOLLET | EPOLLRDHUP,
+    .flags = EPOLLIN | EPOLLET,
     .app_ctx = ctx,
     .free_app_ctx = algo_main_destroy,
     .cb = main_on_udp_read,
@@ -234,7 +236,7 @@ void algo_main_init(struct evt_core_ctx* evt, struct algo_params* ap) {
 
   struct evt_core_cat tcp_write = {
     .name = "tcp-write",
-    .flags = EPOLLOUT | EPOLLET | EPOLLRDHUP,
+    .flags = EPOLLOUT | EPOLLET | EPOLLHUP,
     .app_ctx = ctx,
     .free_app_ctx = algo_main_destroy,
     .cb = main_on_tcp_write,